HAL Id: hal-01388054
https://hal.archives-ouvertes.fr/hal-01388054v3
Preprint submitted on 2 Feb 2018
HAL is a multi-disciplinary open access archive for the deposit and dissemination of sci- entific research documents, whether they are pub- lished or not. The documents may come from teaching and research institutions in France or abroad, or from public or private research centers.
L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.
Darboux integrable system with a triple point and pseudo-abelian integrals
Aymen Braghtha
To cite this version:
Aymen Braghtha. Darboux integrable system with a triple point and pseudo-abelian integrals . 2016.
�hal-01388054v3�
DARBOUX INTEGRABLE SYSTEM WITH A TRIPLE POINT AND PSEUDO-ABELIAN INTEGRALS
AYMEN BRAGHTHA
ABSTRACT. We study pseudo-abelian integrals associated with polynomial perturbations of Dar- boux integrable system with a triple point. Under some assumptions we prove the local boundedness of the number of their zeros. Assuming that this is the only non-genericity, we prove that the num- ber of zeros of the corresponding pseudo-abelian integrals is bounded uniformly for nearby Darboux integrable foliations.
1. INTRODUCTION
Abelian integrals are integralsI(h) = R
γ(h)ηof polynomial or rational one-forms along cycles γ(h)∈H1(H−1(h)), H ∈C[x, y]. Abelian integrals appear as principal part of Poincar´e displace- ment function of the perturbationdH+εηalongγ(h). Their zeros are related to limit cycles born in the perturbation. In [11, 7], Varchenko and Khovanskiii prove the following result
Theorem 1.1. There exist a uniform bound, depending only on the degree of H and η, for the number of real zeros of Abelian integrals.
Arnold posed with insistence the analogous problem for more general polynomial perturbation of integrable systems. In particular, for perturbation of systems having a Darboux first integrals
H = Q
Piαi, Pi ∈ C[x, y]. Then instead of Abelian integrals, one encounters pseudo-Abelian integrals. Pseudo-Abelian integrals naturally appear in generalizations to classes bigger than the Hamiltonian class. The simplest case is the case of Darboux integrable planar foliations F = {MdHH = 0}, where H = Q
Piαi is a product of real powers of bivariate polynomials and M = QPiis the so-called integrating factor, so the formω=MdHH is a polynomial one-form. Poincar´e- Pontryagin criterium claims that in the regular part the limit cycles of the perturbed foliations F ={ω+η= 0}can be born only from those cyclesδof foliationFfor which the integralR
δ η M
vanishes. This integral is called the pseudo-Abelian integral, and currently the main open question is the existence of a uniform bound on the number of its zeros.
2. FORMULATION AND MAIN RESULT
This paper is a part of program of Bobie´nski, Mardeˇsi´c and Novikov to extend the Varchenko- Khovanskii theorem from Abelian integrals to pseudo-Abelian integrals. After studying the generic cases [2, 9] and some nongeneric cases [1, 3, 4], in this work we study a nongeneric case. Let
Date: December 27, 2017.
2010Mathematics Subject Classification. 34C07, 34C08.
Key words and phrases. Abelian integrals, Limit cycles, Darboux integrability.
1
F ={MdHH = 0}denote a foliation with a triple point (assume it to be at the origin), where H =P0
k
Y
i=1
Pii, M =P0
k
Y
i=1
Pi, Pi ∈R[x, y], , i >0,
so that the level curves{P0 = 0},{P1 = 0}and{P2 = 0}intersect transversally two by two at the origin which is the only triple point. LetFλ = {MλdHHλ
λ = 0}be a foliation unfolding F, Mλ an integrating factor, where
Hλ =Pλ
k
Y
i=1
Pii, Mλ =Pλ
k
Y
i=1
Pi. (1)
Generically, a triple point unfolds into three saddles-type singular pointspλ0, pλ1 andpλ2 correspond to the transversal intersections of level curves{P1 = 0}and{Pλ = 0},{P1 = 0}and{P2 = 0}, and{P2 = 0}and {Pλ = 0}. Here also appears a centerpcλ in the triangular region bounded by these levels curves- see Figure 1.
Pi= 0 {Pλ= 0}
Pj= 0 P1= 0
P2= 0
𝑝1𝜆
𝑝2𝜆
𝑝0𝜆 𝑝𝜆𝑐
γ(λ, h)
FIGURE 1. The phase portrait ofHλ
2
Assume that the system MλdHλ
Hλ has a family of cycles γ(λ, h) ⊂ Hλ−1(h) filling a connected component ofR2\ {PλP1· · ·Pk = 0}, which we denote byD. The period annulusDis bounded by{Pλ = 0},{P1 = 0},{P2 = 0}and some separatrices{Pi = 0}, i= 3,· · · , k.
Consider a polynomial perturbation of the systemMλdHHλ
λ. MλdHλ
Hλ +κη, η=Rdx+Sdy, R, S ∈R[x, y], κ >0.
The linear part in perturbation parameterκof the Poincar´e first return map is given by the pseudo- Abelian integral
I(λ, h) = Z
γ(λ,h)
η
Mλ. (2)
We prove the existence of a local bound for the number of zeros of pseudo-Abelian integrals.
LetΩk(λ, n0, n1,· · · , nk;n)be the following, finite dimensional space of polynomial system with Darboux type first integral:
Ωk(λ, n0, n1,· · · , nk;n) := {MλdHλ
Hλ +κη:Hλ =Pλ
k
Y
i=1
Pii,
degPλ ≤n0,degPi ≤ni,deg(R, S)≤n}, where Mλ = PλQk
i=1Pi. The parameters of the space Ωk(λ, n0, n1,· · · , nk;n) are positive ex- ponents(, 1,· · · , k), coefficients of polynomialsPλ, Piand coefficients of the polynomials per- turbation(R, S). We distinguish an open subsetΩk0(λ, n0, n1,· · · , nk;n)defined by the condition that the unperturbed system (withκ = 0) admits a period annulusDand the following genericity assumptions are satisfied
• A1. The Darboux first integralHλ is regular at infinity.
• A2. ∂P∂λλ|(0,0,0) 6= 0.
• A3. The level curves{Pi = 0}, i= 3,· · ·, kare smooth and intersect transversally two by two. Morever, there are not triple intersection points, fori= 3, . . . , k.
• A4. ηvanishes to the order≥4at(x, y) = (0,0)(technical condition)
Theorem 2.1. LetΩ0 ∈ Ωk0(λ, n0, n1,· · · , nk;n). Assume that the pseudo-Abelian integral IΩ0 associated to Ω0 in (2) is not identically zero. Consider the pseudo-Abelian integrals associated to Ω ∈ Ωk0(λ, n0, n1,· · · , nk;n) in a neighborhood of Ω0. Then, there exists an upper bound Z(Ω0;k, n0,· · · , nk;n)for the number of zeros of the pseudo-Abelian integrals associated toΩ∈ Ωk0(λ, n0, n1,· · · , nk;n)close toΩ0.
The main problem is that the Darboux integrable systemsMλdHHλ
λ have a small nest of cycles which shrinks to the origine(0,0)asλ→0, i.e. the centerpλc generates a possible ramification points of I(λ, h)located on a circle whose radius is of order|λ|+1+2. Thus, we can no directly repeat the argument from [2] to get anλ-independent estimate.
3. CONNECTION WITHBOBIENSKI´ ’S RESULT
In [1], the author considers a polynomial one-formω=MdH
H having a Darboux first integral:
H = (x−)aP
k
Y
i=1
Piai, M = (x−)P
k
Y
i=1
Pi (integrating factor),
3
whereP, Pi ∈R[x, y], a, ai ∈R+andis a sufficiently small parameter. He imposes the following genericity assumptions
(1) The Darboux first integralHis regular at infinity.
(2) For = 0, the polycycleγ(0,0)consists of the edges γi(0,0)contained in a smooth part of the level curve Pi−1(0). Any vertex pij, except (0,0), corresponds to the transversal inetrsection of level curvesPi−1(0)andPj−1(0).
(3) The polynomialP has a critical point of Morse type(0,0), i.e. P(x, y) =y2−x2+h.o.t.
Other polynomialsPj satisfyPj(0,0)6= 0, j = 1, . . . , k.
He considers a small polynomial perturbationω =MdHH
ω,ε=ω+εη, η =Rdx+Sdy, R, S ∈R[x, y].
The linear part in perturbation parameterεof the Poincar´e first return map is given by the pseudo- Abelian integrals
I(, h) = Z
γ(,h)
η
M, γ(, h)⊂H−1(h).
Under the above genericity assumptions, Bobie´nski proves the existence of a local upper bound for the number of zeros of corresponding pseudo-Abelian integrals I(, h). However, his study does not give the existence of a uniform bound in afullneighborhood of the parameter space(, h). The precise sense is clear after blow-up. Bobie´nski’s argument works well when one is studying the integral along cycles contained in a neighborhood of the big polycycleδ=δ01tδ02tδ03tδ−t δ+tδ3t · · · tδk, see Figure 2. It corresponds to a sector in the(, h)-space. However, in order to have a study in a full neighborhood of (0,0) in the product space(, h), one must study also cycles landing on the exceptional level on a polycycle made of the same polycyle in the equatorial plane, but completed by any one of the continuous family of curvest = a+2h joining two saddlesp+and p−, see Figure 2. The study of the cycles in this sector is not done in Bobie´nski’s paper.
To complete the proof, which is the main result of this present paper, we assume that the pertur- bative one-formη vanishes to the order≥4at(x, y) = (0,0)(technical condition), but our study covers a full neighborhood of the polycycle in the product space with space of parameters. Also, in this paper Darboux first integral is more general, i.e.1 6=2. We prove the main result in full gen- erality by blowing-up the whole familyFλand subsequently carefully piece-wise investigating the result combining Petrov’s technique [10], general Gabrielov type results and topological methods.
4. RECTIFYING OF DARBOUX FIRST INTEGRAL
Let us a establish a local normal form near the triple point (0,0,0) for the unfolding of the degenerate polycycleP0....Pk = 0.
Proposition 4.1. Under the above assumptions A2,A3, there exists a local analytic coordinate system(x, y, λ)at(0,0,0)such thatHλ takes the form
Hλ = (x−λ)(y−x)+(y+x)−∆, (3) where∆is an analytic unity function∆(0,0,0)6= 0.
Proof. There exists an analytic coordinate system(x, y)at(0,0)such thatP1(x, y) =y∆1, P2(x, y) = x∆2 and Qk
i=3Pii = ∆3, where ∆1,∆2,∆3 are unities. In these coordinate and by Weier- strass preparation theorem we have Pλ = (x −f(y, λ))∆4 where ∆4 is a unit, ∂f∂λ(0,0) 6= 0
4
and ∂f∂y(0,0) 6= 0. A second application of Weierstrass preparation theorem allows us to write f(y, λ) = (y+g0(λ))∆5, where∆5is a unity function and ∂g∂λ0(0)6= 0. Now we putx˜= ∆x
5. Then
˜
x∆5+ (y+g0(λ))∆5 = (˜x−y−g0(λ))∆5.
Finally, Pλ = (˜x−y−λ)∆˜ 4∆5, where λ˜ = g0(λ). The normal form (3) can be obtained by a
linear rotation on(˜x, y).
5. DARBOUX INTEGRABLE FOLIATION
Consider two Darboux integrable foliations of dimension two in dimension three space with coordinates(x, y, λ)
F1 ={MλdHλ
Hλ = 0}, F2 ={dλ= 0}.
LetF = {MλdHHλ
λ ∧dλ = 0}be the foliation of dimension one in a space of dimension three with coordinates (x, y, λ) which is given by the intersection of the leaves of F1 and F2. This foliation has a non-elementary singularity at the origin(0,0,0)which is the triple point. To reduce this singularity, we make the blowing-up of the triple point of the family in the product space (x, y, λ)of phase and parameter spaces. The family blowing-ups were introduced by Denkowska and Roussarie in [5].
5.1. Blowing-up the triple point. The blowing up σwith center at the origin (the triple point) is the projection ontoC3of a spaceW that is obtained by replacing the origin by the projective space CP2of all lines throught the origin:
W ={(p, q)∈C3×CP2 :p∈q}
andσ :W → C3is defined by σ(p, q) = p. Outside the origin, a pointpbelongs to a unique line q, butσ−1(0) = CP2 which is called the exceptional divisor. If we write pin terms of the affine ccordinatesp = (p1, p2, p3), andqin the corresponding homogeneous coordinatesq = [q1, q2, q3], then the relationp∈qtranslates into the system of equationspiqj =pjqi, fori, j = 1,2,3.
The projective spaceCP2is covered by three canonical charts: W1 ={x6= 0}with coordinates (Y1, E1),W2 ={y6= 0}with coordinates(X2, E2)andW3 ={λ 6= 0}with coordinates(X3, Y3).
W1, W2 and W3 define canonical charts on W, with coordinates (X1, Y1, E1),(X2, Y2, E2) and (X3, Y3, E3)respectively. In these coordinates,σis given by the formulas:
σ1 =σ|W1 :x=X1 y=X1Y1 λ=E1X1, (4) σ2 =σ|W2 :x=X2Y2 y=Y2 λ=E2Y2, (5) σ3 =σ|W3 :x=X3E3 y=Y3E3 λ=E3. (6) We apply the blowing-upσto the one-dimensional foliation F on the three-dimensional space with coordinates(x, y, λ)given by intersection ofMλdHHλ
λ = 0anddλ = 0. Letσ−1F be the lift of the foliationF to the complement of the exceptional divisorCP2.
Proposition 5.1. This foliationσ−1F extends in a unique way to a holomorphic singular foliation σ∗F onW which we call the blow-up of the original dimension-one foliationFby the mapσ. The foliationσ∗F is regular outside of the preimage of the hypersurface{H(x, y, λ) = 0, λ= 0}.
5
5.1.1. Singularities of σ∗F. The strict transform of the period annulus Dlies completely within the chart W1. Then, we concentrate our study uniquely on this chart. Let σ∗1F be the restriction of the blown-up foliation σ∗F to the chart W1. We have σ is a biholomorphism outside CP2 (exceptional divisor), all singularities of σ1∗F on W1 \ {X1 = 0} correspond to singularities of F. Thus, it suffices to compute the singularities ofσ1∗F on the exceptional divisor {X1 = 0}.
Explicitly, near the exceptional divisor{X1 = 0}, the blown-up foliationσ∗1F is given by two first integralsλ=X1E1and λha =G, where
G=E1a(1−E1)−(Y1−1)−+(Y1+ 1)−−∆˜−1,
∆˜ is unit of the form∆ =˜ c+X1f, f is a holomorphic function anda =+++−. Consider the two-dimensional squareQ⊂CP2 with verticesp+, p−, q+andq−. All levels curves{G= λha} insideQcorrespond to values of λha ∈[0,+∞].
Proposition 5.2. The singularities ofσ1∗F on the exceptional divisorCP2 are located at the points p+ = (0,1,0), p− = (0,−1,0), q+ = (0,1,1)andq− = (0,−1,1). All these singular points are linearisable saddles, with eigenvaluesµ+ = (+,−a,−−), µ− = (−−, a, −), ν+ = (0,−, +) andν− = (0,−, −)respectively.
Proof. Let us compute the eigenvalues at p+, p−, q+ and q−. Near p+ the foliation σ1∗F is given by the two first integralsh =XaY+ andλ = XE(we make a convenient variables change). By Hartman-Grobman theorem the vector field generating the foliationσ∗1F is given by
X(X, Y, E) = µ+1X ∂
∂X +µ+2Y ∂
∂Y +µ+3E ∂
∂E,
such that the vector<(µ+1, µ+2, µ+3),(a, +,0)>= 0and<(µ+1, µ+2, µ+3),(1,0,1)>= 0. By short computation, we obtain
X(X, Y, E) = +X ∂
∂X −aY ∂
∂Y −+E ∂
∂E.
Similar computation shows that there are local coordinates nearq+ in which the vector field gen- erating the foliation is given by
Y(X, Y, E) = −Y ∂
∂Y ++E ∂
∂E.
6. PROOF OF THEOREM 2.1
Let us fix some useful notations. Lett = λha, wherea =+−++andQ ⊂CP2 is the two- dimensional square with verticesp+, p−, q+andq−. All levels curves{G=t}insideQcorrespond to values oft∈[0,+∞], where
G=E1a(1−E1)−(Y1−1)−+(Y1+ 1)−−∆˜−1.
6.1. Polycycles, relative cycles and normal form. The important advantage of making the blowing- up σ is to obtain a family of hyperbolic polycycles, i.e. at each intersection of two consecutive curves we have a saddle point. We consider the family of hyperbolic polycyclesδ
δ = σ−11 (γ(0,0)\(0,0,0))∪(Q∩ {G=t})R
, t∈[0,+∞], (7)
where(. . .)Rdenotes the real part of a complex analytic set.
6
q+
q-
p+
p-
03
01
02
04
0
+
-
i
(
FIGURE 2. The foliationσ∗1F
6.1.1. Polycycles. Let δ0 (edge) be the real part of the complex analytic set {X1 = E1 = 0}
joining two saddlesp− andp+,δ01 be the real part of the complex analytic set{X1 = 0, Y1 = 1}
joining the two saddlesp+andq+,δ02be the real part of complex analytic set{X1 = 0, Y1 =−1}
joining two saddles p− and q−, δ03 be the real part of complex analytic set {X1 = E1 = 0}
joining two saddles q− and q+, δ04 be the real part of complex analytic set {X1 = 0, G = t}
joining two saddlesp−andp+, δ+ be the real part of the complex analytic set{Y1 = 1, E1 = 0}
joining two saddlesp1 andp3, δ−be the real part of the complex analytic set{Y1 = −1, E1 = 0}
joining the two saddles p− and pk and finally let δi be the real part of the complex analytic set {σ1∗Pi = 0, E1 = 0}, i= 3,· · · , kjoining the two saddlespmandpn-see Figure 2.
Let0≤m < M. Then, we can decompose the family of hyperbolic polycyclesδas follows:
(1) Ift∈[0, m[, we haveδ=δ0tδ−tδ+tδ3t · · · tδk.
7
(2) Ift∈[m2,2M], we haveδ =δ04tδ−tδ+tδ3t · · · tδk;
(3) Ift∈[M,+∞], we haveδ =δ01tδ02tδ03tδ−tδ+tδ3t · · · tδk.
6.1.2. Relative cycles. Letp1,· · · , pkbe the saddles points of the foliation σ∗1F which lie on the polycycle δ. Let δ0 = δ \ {p1,· · · , pk}. Choose a family of analytic transversals (of complex dimension two)Σx, through each point pointxinδ0. We can define a relative cycleγ (the part of the cycleδ(λ, t)) by a initial condition (starting pointx) and a end pointy =γ∩Σywhich is going fromΣx. Concretly, we consider the relative cyclesδ0(λ, t)corresponding to the edgeδ0,δ01(λ, t) corresponding to the edgeδ01, δ02(λ, t)corresponding to the edgeδ02, δ03(λ, t)corresponding to the edge δ03, δ04(λ, t) corresponding to the edge δ04, δ+(λ, t) corresponding to the edge δ+, the relative cycleδ−(λ, t)corresponding to the edge δ− and the relative cyclesδi(λ, t)corresponding to the edgeδi, i= 3,· · · , k- see Figure 3.
q+
q-
p+
p-
03(
01(
0 (
+ (
- (
i (
FIGURE 3. The relative cycles
8
6.1.3. Normal form coordinates near the polycycles. Now we obtain normal forms in neighbor- hoods of each edge of polycyles
Proposition 6.1. (1) There exist a local chart(U0,(X, Y, E))⊂W defined in a neighborhood ofδ0 such that the blown-up foliationσ1∗F is given by two first integrals
λ=XE, t = (Y −1)−+(Y + 1)−+Ea.
(2) There exist a local chart(U01,(X, Y, E))⊂W defined in a neighborhood ofδ01such that the blown-up foliationσ1∗F is given by two first integrals
λ =XE, t=Ea(1−E)−Y−+.
(3) There exist a local chart(U02,(X, Y, E))⊂W defined in a neighborhood ofδ02such that the blown-up foliationσ1∗F is given by two first integrals
λ=XE, t=Ea(1−E)−Y−−.
(4) There exist a local chart(U03,(X, Y, E))⊂W defined in a neighborhood ofδ03such that the blown-up foliationσ1∗F is given by two first integrals
λ=X, t=E−(Y −1)−+(Y + 1)−−.
(5) There exist a local chart(U+,(X, Y, E)) ⊂W defined in a neighborhood ofδ+ such that the blown-up foliationσ1∗F is given by two first integrals
λ=XE, t=Y+(1 +X)iEa.
(6) There exist a local chart(U−,(X, Y, E)) ⊂W defined in a neighborhood ofδ− such that the blown-up foliationσ1∗F is given by two first integrals
λ=XE, t=Y−(1 +X)iEa.
(7) There exist a local chart(Ui,(X, Y, E))⊂W, i = 3,· · · , k, defined in a neighborhood of δi such that the blown-up foliationσ1∗F is given by two first integrals
λ =E, t=Yi(1−X)i+1Xi−1 =t.
6.2. Proof of Theorem 2.1. Let δ be a polycycle. Let δ(λ, t) = σ−1(γ(λ, h)) ⊂ W (dashed cycle, see Figure 2) be the pull-back of the cycle γ(λ, h) by the blowing-up map and δ be its corresponding polycycle. We define the integral
J(λ, t) = Z
δ(λ,t)
σ1∗ η
Mλ. (8)
This integral is considered as the pull-back of the pseudo-abelian integralsI(λ, h)by the blowing- upσ1, i.e. J(λ, t) =σ1∗I(λ, t). The proof of Theorem 2.1 is reduced to the proof of the following theorem
Theorem 6.2. Letε > 0be sufficiently small. Then, for all|λ| < εthe number#{t ∈ [0,+∞] : J(λ, t) = 0}is locally bounded.
9
6.2.1. Variation operator. Firstly, let us recall some definitions, notation and general results. They will be useful later.
Definition 6.3. Given any multivalued function J defined in a neighborhood of the origin in C i.e. a holomorphic function defined on the universal coveringCf∗ of C∗. We define the rescaled monodromy as
Mon(t,α)J(t) = J(teiπα).
The variation is given as the difference between the counterclockwise and clockwise continuation Var(t,α)J(t) =Mon(t,α)J(t)− Mon(t,−α)J(t)
=J(teiπα)−J(te−iπα).
Definition 6.4. Let J be a multivalued function in two variables λ and t defined in universal coveringC2\ {λt^= 0}ofC2\ {λt= 0}. We define the mixed variation as
Var(λ,β)◦ Var(t,α)J(λ, t) =Var(λ,β)(J(λ, teiπα)−J(λ, te−iπα)) =
J(λeiπβ, teiπα)−J(λe−iπβ, teiπα)−J(λeiπβ, te−iπα) +J(λe−iπβ, te−iπα).
Lemma 6.5. The variationsVar(λ,β) andVar(t,α)commute
Var(λ,β)◦ Var(t,α) =Var(t,α)◦ Var(λ,β).
Proof. The proof is a consequence of the monodromy theorem which says that: Ifγ1, γ2 are ho- motopic paths inC2\ {λt= 0}, thenψγ1 =ψγ2 whereψγ1 =Monγ1ψ andψγ2 =Monγ2ψ. We consider
γ1(θ, φ) = (λ(θ, φ), t(θ, φ)) = λ, teiθ
θ∈[0,α]t λeiφ, teiα
φ∈[0,β], γ2(θ, φ) = (λ(θ, φ), t(θ, φ)) = λeiφ, t
φ∈[0,β]t λeiβ, teiθ
θ∈[0,α].
The paths γ1 andγ2 are homotopic and this implies that ψ(λeiαπ, teiβπ)can be defined either as ψγ1 orψγ2. The same argument holds for the other germsψ(λe−iαπ, teiβπ),
ψ(λeiαπ, te−iβπ)andψ(λe−iαπ, te−iβπ).
6.2.2. Analytic properties. The integralJ(λ, t)has an analytic extension to the complex argument t(respλ). This is a multivalued function with unique branch pointt = 0(respλ = 0). As in [2], the key of the proof of Theorem 6.2 is the following.
Proposition 6.6. The integralJ(λ, t)satisfies the following iterated rescaled variations equation Var(t,α1)◦. . .◦ Var(t,αk)J(λ, t) = 0. (9) whereαiare polynomials functions in, +, −, 3,· · · , k.
Proof. Let us fixλ. We choose a hyperbolic polycycle δof the family (6.1). As in [2], using the different charts of Proposition 6.1 and partition of unity multiplying the blown-up one formσ1∗Mη
λ
we can consider semilocal problem with a relative cycleδi(λ, t)(part of cycleδ(λ, t)) close to one edge (i-th edge) of the polycycle. LetδiCbe the complexification of the reali-th edge joining the singular pointspi−1, pi+1(saddles).
10
(1) Ifαi−1 6=αi+1 (generic case), we have
Var(t,αi−1)◦ Var(t,αi+1)δi(λ, t) =[γi−1^, γi+1], (10) where[γi−1^, γi+1]is a complex (closed) cycle obtained as a lift of the commutator[γi−1, γi+1], whereγi−1 and γi+1 are paths in δiC\ {pi−1, pi+1} turning once counterclockwise around pi−1 andpi+1. This lifting[γi−1^, γi+1]vanishes by making third variation, i.e.
Var(t,αi)[γi−1^, γi+1]≡0. (11) (2) Ifαi−1 =αi+1 (resonant case), we have
Var(t,αi−1)δi(λ, t) =γi−1^γi+1−1, (12) where γ^i−1γi+1−1 is a complex (closed) cycle obtained as a lift of the figure eight loop γi−1γi+1−1. This liftingγ^i−1γi+1−1 vanishes by making second variation, i.e.
Var(t,αi)γ^i−1γi+1−1 ≡0. (13) Finally, the variations commute so
Var(t,α1)◦. . .◦ Var(t,αk)δ(λ, t) = 0. (14) This argument is independent of the choice of polycycle, i.e. it holds for any hyperbolic polycycle
δof family (7).
Proposition 6.7. The rescaled variation with respect to λof the integral J(λ, t) is an integral of the formσ∗1Mη
λ along the figure eight loop
Var(λ,1)J(λ, t) = Z
eight loop
σ∗1 η
Mλ. (15)
Proof. In the local chart (U0,(X, Y, E)) of Proposition 6.1, the blown-up foliationσ1∗F is given by the two first integralsλ=XE andt=Ea(Y −1)−+(Y + 1)−−. Letγ+andγ−be two paths inYCturning counterclokwise aroundp+andp−which are parametrized by
ρ± :θ∈[0,2π]7→
X(θ, λ, t)
Y(θ) = ±1 +εeiθ E(θ, λ, t)
.
Then, we have
F±(λ, t) = Z
γ±
σ1∗ η Mλ =
Z 2π 0
ρ∗±σ∗1 η Mλdθ.
Also, we define two functions F1(λ, t) =
Z
`−
σ1∗ η
Mλ, F2(λ, t) = Z
`+
σ∗1 η Mλ,
where`−= [1−ε,−1 +ε]and`+ = [−1 +ε,1−ε](segments). So, we obtain Var(λ,1)J(λ, t) = F−(λ, t) +F2(λ, t) +F+(λ, t) +F1(λ, t) =
Z
γ+`^−γ−`+
σ∗1 η Mλ,
11
whereγ+^`−γ−`+is a closed path obtained as a lift of the pathγ+`−γ−`+which is contained inYC
and homotopic to a figure eight loop.
Corollary 6.8. Near the ramification pointλ= 0, the functionJ(λ, t)admits the expansion J(λ, t) = J1(λ, t) +J2(λ, t) logλ, (16) whereJ1(., t)is meromorphic andJ2(λ, t) =Var(λ,1)J(λ, t).
6.2.3. Proof of Theorem 6.2. The integralJ(λ, t) = R
δ(λ,t)σ∗1Mη
λ can be analytically continued to the universal coverC2\ {λt^= 0}ofC2\{λt= 0}. To estimate the number of zeros of the integral J(λ, t) =R
δ(λ,t)σ∗1Mη
λ we apply the argument principle.
Let us introduce some definitions which will be useful later.
Definition 6.9. Let f : Rn× R → R. We shall say that f is a logarithmico-analytic function (LA-function) of type`in variableyif it has the following form
f(x, y) = F(f1(x, y), . . . , fm(x, y),logfm+1(x, y), . . . , fm+r(x, y)), whereF is a global sub-analytic function andfiare a LA-functions of type`−1iny.
Definition 6.10. A logarithmico-exponential function (LE-function) is a finite composition of global sub-analytic functions, exponentials and logarithms.
Let ∂Ω be the boundary of a complex domain Ωwhich consists of a big circular arc CR1 = {|t| = R1,|argt| ≤ απ}, a two segmentsC± = {r1 ≤ |t| ≤ R1,|argt| = ±απ}and the small circular arcCr1 ={|t|=r1,|argt| ≤απ}-see Figure 4.
The argument principle says that
#{t ∈Ω :J(λ, t) = 0} ≤ 1
2π∆ arg∂ΩJ = 1
2π(∆ argCR
1 J+ ∆ argC±J+ ∆ argCr
1 J).
(1) The boundedness of the increment of argument ∆ argC
R1 J. By Gabrielov’s theorem [6], the increment of the argument∆ argCr
1 J is uniformly bounded.
(2) The boundedness of the increment of argument∆ argC±J. Letα ∈ {α1,· · · , αk}. We use Schwartz’s principle
Im(J(λ, .))|C± =∓2iVar(t,α)J(λ, t).
So ∆ argC±J ≤ #{t : Im(J(λ, .)) = 0} = #{t : Var(t,α)J(λ, t) = 0}. Moreover, the variations commute so
Var(t,α1)◦ · · · ◦ Var(t,α)◦ · · · ◦ Var(t,αk)J(λ, t) = Var(t,α1)◦ · · · ◦ Var(t,αk)(Var(t,α)J(λ, t)) = 0.
Then, near the ramification point t = 0, the function Var(t,α)J(λ, t) can be written as follows
Var(t,α)J(λ, t) =F(eαα1logt, . . . , eαkα logt,logλ) (17) whereF is a meromorphic function. The functionVar(t,α)J(λ, t)is a LA-function of type 1 in variableλ. Then, by Lion-Rolin preparation theorem [8] this function has the following form
Var(t,α)J(λ, t) =λq00λq11G(t)U(t, λ0, λ1), U(0,0,0)6= 0 (18)
12
Re(t) Im(t)
0
C+
C-
CR1
Cr1 𝛼𝜋
FIGURE 4. The contour∂Ω
withλ0 =λ−θ0(t), λ1 = logλ0 −θ1(t), whereθ0, θ1,GandUare LE-functions. As the number of zeros of a LE-function is bounded, so #{t : Var(t,α)J(λ, t) = 0}is uniformly bounded inλ.
(3) The boundedness of the increment of argument ∆ argCr
1 J. Consider the following func- tional spaceP
P(m, M;α1, . . . , αk;λ) := {X X
cjln(λ)tαjnlognt, cjln ∈C, m≤αjn≤M,0≤l≤k}.
Proposition 6.11. We haveJ2(λ, t) = Var(λ,1)J(λ, t) = O(λµ) uniformly int, for some constantµ >0.
Proof. Using the assumptionA4, we haveσ1∗Mη
λ isO(X1), we conclude that, for all closed paths of finite lenght` < ∞contained in a sufficiently small neighborhood of the excep- tional divisor{X1 = 0}. Since J2(λ, t) =Var(λ,1)J(λ, t)is the integrations ofσ1∗Mη
λ over the lift of the eight figure on{X1 = 0, G = t}, we conclude thatX1 = O(λ)on this lift and
J2(λ, t) =Var(λ,1)J(λ, t) =λµtν(1 +. . .), µ >0.
13
Lemma 6.12. The functionsJ1(λ, .), J2(λ, .)are two meromorphic families inλand satisfy the following rescaled variation equation with respect tot
Var(t,α1)◦. . .◦ Var(t,αk)Ji(λ, t) = 0, i= 1,2. (19) Then, there exists a family of meromorphic functionP1(λ, .), P2(λ, .)∈ P(...)such that
|Ji(λ, t)−Pi(λ, t)| ≤C|t|M, uniformly in λ, i= 1,2
and J2(λ, t)−P2(λ, t) = O(λµ), µ > 0uniformly int and(J2(λ, t)−P2(λ, t)) logλ = O(λµlogλ). MoreoverJ(λ, t)6= 0. Then for sufficiently bigM:P1(λ, t)+P2(λ, t) logλ6=
0.
Proof. Using the linearity of the variation operatorVar, equations (9) and (16), we have Var(t,α1)◦. . .◦ Var(t,αk)Ji(λ, t) = 0, i= 1,2.
Lemma 4.8 from [2] yields that there exists an analytic (a priori meromorphic) families of functionsPi(λ, .)∈ P(. . .)such that|Ji(λ, t)−Pi(λ, t)| ≤C|t|M, uniformly inλ.
To estimate the limit of the increment of argument∆ argCr
1J(λ, t)along the small cir- cular arc Cr1: limr1→0∆ argCr
1 J, we investigate the leading term of J(λ, t) at t = 0.
By Lemma 6.12 we have J1(λ, t) +J2(λ, t) logλ−(P1(λ, t) +P2(λ, t) logλ)is O(tM) uniformly inλ. For each element of parameters space, we can choose the leading termP of P1(λ, t) +P2(λ, t) logλ. By Gabrielov’s theorem, the increment of argument of P is bounded.
ACKNOWLEDGMENTS
It is pleasure to thank Pavao Mardeˇsi´c (Universit´e de Bourgogne-Dijon) and Daniel Panazzolo (Universit´e de Haute Alsace-Mulhouse) for their constants suport in this work.
REFERENCES
[1] Bobie´nski, Marcin Pseudo-Abelian integrals along Darboux cyclesa codimension one case. J. Differential Equa- tions 246 (2009), no. 3, 1264-1273.
[2] Bobie´nski, Marcin; Mardeˇsi´c, PavaoPseudo-Abelian integrals along Darboux cycles. Proc. Lond. Math. Soc. (3) 97 (2008), no. 3, 669-688.
[3] Bobie´nski, Marcin; Mardeˇsi´c, Pavao; Novikov, DmitryPseudo-Abelian integrals: unfolding generic exponential.
J. Differential Equations 247 (2009), no. 12, 3357-3376.
[4] Bobie´nski, Marcin; Mardeˇsi´c, Pavao; Novikov, DmitryPseudo-Abelian integrals on slow-fast Darboux systems.
Ann. Inst. Fourier (Grenoble), 2013
[5] Denkowska, Zofia; Roussarie, Robert; A method of desingularization for analytic two-dimensional vector field families. Bol. Soc. Brasil. Mat. (N.S.) 22 (1991), no. 1, 93-126.
[6] Gabri`elov, A. MProjections of semianalytic sets. (Russian) Funktsional. Anal. i Priloˇzen. 2 1968 no. 4, 18-30 [7] Khovanskii, A, G.Real analytic manifolds with property of finitness, and complex abelian integrals, Funktsional.
Ana. i Priloˇzen. 18 (2) (1984) 40-50.
[8] Lion, Jean Marie; Rolin, Jean PhillipeTh´eor`eme de pr´eparation pour les fonctions logarithmico-exponentielles.
Annales de l’institut de Fourier, tome 47, n3 (1997) p.859-884.
[9] Novikov, DmitryOn limit cycles appearing by polynomial perturbation of Darbouxian integrable systems. Geom.
Funct. Anal. 18 (2009), no. 5, 1750-1773.
[10] Petrov, G. S.On the nonoscillation of elliptic integrals. (Russian) Funktsional. Anal. i Prilozhen. 31 (1997), no.
4, 47–51, 95; translation in Funct. Anal. Appl. 31 (1997), no. 4, 262-265 (1998).
[11] Varchenko, A. N.Estimation of the number of zeros of an abelian integral depending on a parameter, and limit cycles. (Russian) Funktsional. Anal. i Priloˇzen. 18 (1984), no. 2, 14-25.
14
UNIVERSITE DE´ BOURGOGNE, INSTITUT DEMATHEMATIQUES DE´ BOURGOGNE, U.M.R. 5584DUC.N.R.S., B.P. 47870, 21078 DIJON CEDEX- FRANCE.
E-mail address:aymenbraghtha@yahoo.fr
15